Feature extraction and recognition of wall impact noise

Recognition of the impact noises emitted by electrical-drill or hammer is significant in such applications as building sound event monitoring system. The methods based on spectral features have been proved to have poor performance, therefore, in this paper a novel temporal feature has been proposed considering the similarity and volatility in frequency domain and the difference in time domain between various types of noises. When using measured data to testify the performance, it can be found that the recognition accuracy of the temporal features is better than that of the popular spectral features such as MFCC and the feature is noise robust.